SUBMITTED TO IEEE SIGNAL PROCESSING …users.ece.utexas.edu/~bevans/papers/2012/powerline/... ·...

SUBMITTED TO IEEE SIGNAL PROCESSING MAGAZINE 1

Local Utility Powerline Communications in

the 3-500 kHz Band:

Channel Impairments, Noise, and StandardsMarcel Nassar, Student Member, IEEE, Jing Lin, Student Member, IEEE,

Yousof Mortazavi, Student Member, IEEE, Anand Dabak, Senior Member, IEEE,

Il Han Kim, Member, IEEE, and Brian L. Evans, Fellow, IEEE

Abstract

Future Smart Grid systems will intelligently monitor and control energy flows in order to

improve the efficiency and reliability of power delivery. This monitoring and control requires

low-delay, highly reliable communication between customers, local utilities and regional utilities.

A vital part of future Smart Grids is the two-way communication links between smart meters

at the customer sites and a (decentralized) command and control center operated by the local

utility. To enable these two-way communication links, narrowband powerline communication

(PLC) systems operating in the 3-500 kHz band are attractive because they can be deployed

over existing outdoor power lines. Power lines, however, have traditionally been designed for

one-directional power delivery and remain hostile environments for communication signal prop-

agation. In this article, we review signal processing approaches to model channel impairments

and impulsive noise and mitigate their effects in narrowband PLC systems. We examine ways

to improve the communication performance based on current and emerging standards.

Index Terms

Channel modeling, impulsive noise, cyclostationary noise, narrowband PLC, power grid,

power line channel, power line communications, Smart Grid.

M. Nassar, J. Lin, Y. Mortazavi, and B. L. Evans are with the Dept. of Electrical and Comp. Eng., The University

of Texas, Austin, Texas USA. (see http://users.ece.utexas.edu/˜bevans/projects/plc/).

A. Dabak and I. Kim are with Texas Instruments Research and Development Center in Dallas, Texas USA.

Manuscript received September 6, 2011, resubmitted on January 20, 2012.


I. INTRODUCTION

Traditionally, a local utility has had to balance energy demand by local customers with energy

supply. The energy supply came from power plants run by the local utility and/or the regional

utility. Energy and information flowed only in one direction to the end user. In the future, a

two-way flow of energy and information will occur throughout the Smart Grid.

The bidirectional flow of information will help the local utility manage the wide variety of types

of power consumers and generators on the grid to guarantee quality of service while improving

energy efficiency. To enable bidirectional flow of information, powerline communication (PLC)

systems have been developed for use in every part of the grid. PLC systems may be placed in

three bandwidth categories [1] as explained below.

Ultra narrowband (UNB), or very low frequency (VLF), PLC systems operate in the 0.3-3.0

kHz band to provide about 100 bps over ranges up to 150 km. An example, that has been deployed

by many utilities, is the Two-way Automatic Communication System (TWACS) that sends two

bits per mains cycle (i.e. 100 bps in Europe and 120 bps in North America). TWACS applications

include automatic meter reading (AMR), outage detection, and voltage monitoring.

Narrowband (NB), or low frequency (LF), PLC systems operate in the 3-500 kHz band to

deliver a few kbps in the single carrier case and up to 800 kbps in the multicarrier case. Several

standards for commercial building automation, such as BacNet (ISO 16484-5) and LonTalk

(ISO/IEC 14908-3), use single carrier NB-PLC. PRIME1 and G32 are the well-known industry-

developed multicarrier NB-PLC standards. Recently, ITU approved a family of standards that

added G.hnem to the established PRIME and G3. These, in addition to the ongoing IEEE

P1901.2 effort, extend the capabilities of local utilities into device-specific billing and smart

energy management.

Broadband (BB), or high frequency (HF), PLC systems operate in the 1.8-250 MHz band to

provide up to 200 Mbps for home area networks. The IEEE P1901 standard and the approved

ITU G.996x recommendations (G.hn) are the most recent broadband standards. These systems are

based on multicarrier communications. Furthermore, the first MIMO broadband PLC standard,

using multiple available wires, was approved (Dec. 2011) as ITU Recommendation G.9963.

1PoweRline Intelligent Metering Evolution: http://www.prime-alliance.org2http://www.maxim-ic.com/products/powerline/g3-plc


Fig. 1. Network-level system model for local utility powerline communications for US deployment: Meters transmit

usage and load information on the LV access network through the transformer to a MV router. In turn, the MV router

forwards this data to the concentrator that sends it to the utility company over a telecommunication backbone.

In the paper, we discuss narrowband PLC (NB-PLC) for local utility applications. A significant

challenge in improving communication performance is overcoming channel distortion and noise.

Channel distortion effects include the usual suspects of attenuation, frequency selectivity and

fading due to the power transmission lines. They also include the effects of the transformer that

bridges the low-voltage power line at the consumer with the medium voltage line for the local

utility. Impulsive noise may be tens of deciBels higher than the background (thermal) noise. We

model both the PLC channel and noise and show how current and emerging NB-PLC standards

propose to deal with these impairments.

II. SYSTEM MODEL OF LOCAL UTILITY PLC

PLC is envisioned as a platform for various Smart Grid applications. These applications include

real-time monitoring and load balancing, improving grid robustness to disturbances, integrating

alternate energy sources into the grid, and electric car charging and billing. Smart metering is

currently one of the main target applications for PLC. Frequent meter readings of up to one read

per 15 minutes per meter can provide the utility and the end user with better information about

the load usage. The utility can use this information to optimize power generation and perform

load management with finer time granularity. The users, on the other hand, can use this feedback

to adjust their usage patterns to save on energy costs.

The basic system model for a utility PLC network for metering applications is given in Fig. 1.


The meter at the house collects usage information and sends this data across the low-voltage (LV)

access network. Depending on the deployment environment, this information is aggregated at an

LV router and passed through the transformer to an MV router (in Europe), or directly through the

transformer to a medium-voltage (MV) router (in the US as shown in Fig. 1). Consequently, the

MV routers forward the information to a concentrator that is connected to the utility company

through a telecommunication backbone (such as WAN). While a bus topology is depicted in

Fig. 1, other access topologies such as star and ring topologies are possible [2], [3]. Control data

from the utility would flow in the opposite direction toward the customer.

III. CHANNEL MODELS

In PLC systems, the electric grid constitutes the transmission medium. The signal propagates

through electric wires and various discrete components such as transformers. In NB-PLC, the

typical wire length in an LV or MV network is comparable to the propagation wavelength λ.

For example, at f = 500 kHz, the typical wire length is greater or on the order of λ/4 = ν4f ≈

100m, where ν is the signal propagation speed. A TL of length λ/4 will cause a phase shift of

π rads, and may give rise to a π-shifted reflected wave traveling in the opposite direction. As

a result, the signal propagation is dominated by transmission line (TL) effects. Furthermore, the

electrical length of power lines in NB-PLC networks is typically short enough to make the access

impedance, determined by connected loads, dominate the characteristic impedances of the lines3

at the input of PLC modems. These factors result in frequency selective channels with potentially

deep notches that affect the communication performance of PLC systems. Fig. 2 shows examples

of frequency selective responses and their corresponding root-mean-square (RMS) delay spreads

encountered in the field. The RMS delay spread στ,µs is a measure of the extent the channel is

spread in time and is equal to the second central moment of the power delay profile. Additional

loss is introduced when the signal passes through a transformer from the MV to the LV side.

This is shown in Fig. 2 (4), where the channel has significantly higher attenuation than the other

three cases. Next, we introduce two popular approaches for modeling PLC channels.

3The electrical length of a cable is its length expressed as the number of wavelengths of the propagating signal. As

the electrical length tends to ∞, its reflected impedance approaches the characteristic impedance Z0 (f) of the wire

(this is the case in some broadband PLC systems).


50 100 150 200 250 300 350 400 450−80

−70

−60

−50

−40

−30

−20

−10

0

Frequency (kHz)

Cha

nnel

Mag

nitu

de (d

B)

Cha

nnel

Mag

nitu

de (d

B)

50 100 150 200 250 300 350 400 450−55

−50

−45

−40

−35

−30

−25

−20

−15

Frequency (kHz)

50 100 150 200 250 300 350 400 450−50

−45

−40

−35

−30

−25

Frequency (kHz) 50 100 150 200 250 300 350 400 450−100

−90

−80

−70

−60

−50

−40

Frequency (kHz)

(1) Residential Low-Voltage Site (China) RMS Delay Spread = 3.96 f

(2) Industrial Low-Voltage Site (India)RMS Delay Spread = 2.95 t

(3) Industrial Low-Voltage Site (India)RMS Delay Spread = 2.22 g

(4) Residential MV to LV Site (USA)RMS Delay Spread = 5.36 h

Fig. 2. Examples of different PLC channels in frequency domain and their corresponding RMS delay spreads.

A. Multi-Path Modeling

The multi-path model views the channel as an unknown function that needs to be estimated

based on training data; hence, it does not use any prior information such as the network topology

or component connections. Instead, it proposes a parametric model of the channel frequency

response based on TL theory. This modeling approach is useful when the network topology is

not available or an ad hoc deployment is desired. The multi-path model was first proposed in [4]

and later augmented in [5]. It is based on the superposition of signal reflections at various loads

predicted by TL theory. Its parametric form is given by

H (f) =N∑i=1

gie−α(f)li × e−j2πfτi (1)

where N is the number of propagation paths and gi, α (f), li, τi are the gain, the attenuation factor,

the length, and the delay of each path, respectively [5]. This modeling approach is illustrated in

Fig. 3, where a sequence of training symbols is used to estimate the frequency response of the

PLC channel. The accuracy of this model is highly dependent on N , the number of significant

paths, with values between 15 to 44 required for a fit of a 110 m link [5]. The high number of

parameters and the complex parameter estimation procedure are some of the disadvantages of this

approach; however, the primary disadvantage is its inability to predict the channel response based

on the PLC network topology and how it varies given a change in this topology. Furthermore,


Fig. 3. The multi-path model approach: i) training symbols are sent over the channel, ii) the channel parameters are

estimated, and iii) these parameters are used to interpolate the channel response at any frequency range using (1).

it is not clear how the model parameters change from site to site. This limits its applicability in

coverage and deployment analysis for PLC systems.

On the other hand, the parametric form in (1) gives rise to a statistical description of the PLC

channel if the parameters are viewed as random variables across different sites. An experimental

and statistical analysis of such parameters for wireline channels is given in [6], where it is shown

that the channel’s average power gain G has a lognormal distribution. A random variable is

said to be lognormal distributed if its logarithm is normally distributed. This usually arises in

multiplication of i.i.d. random variables as in the channel gain of a PLC channel. Furthermore, a

negative correlation between the RMS delay spread στ,µs and GdB is observed which could have

an impact on the use of ISI-mitigation techniques. This correlation is modeled using a linear

regression fit of the form

log (στ,µs) = θGdB + ζ (2)

where the parameter fits for θ and ζ are provided for multiple practical wireline channels [6].

B. Transmission-Line Modeling

Transmission-line modeling, originally introduced for modeling DSL channels [7], views the

channel as a deterministic quantity that is a function of the network topology and its electrical

components. Two-conductor and multi-conductor transmission line models are two examples of

bottom-up approaches using TL theory [7], [8]. In particular, a powerful approach is to model

the electrical components of the PLC network as 2-port networks (2PN) and use these network

parameters to relate the voltages and currents across the ports. ABCD parameters are particularly

attractive due to the ease of their concatenation for different network topology configurations

(simple matrix multiplications) [7]. Furthermore, using transmission line and circuit theory, the


Fig. 4. The transmission line approach: i) the network topology is mapped to a tree of electrical components described

by their ABCD models, then ii) the channel is estimated using the concatenation of ABCD parameters. A 3D figure of

channel response vs. frequency vs. distance using transformer measurements can be generated and used in deployment

studies. The huge dip in the channel gain is due to crossing the medium to low voltage through the transformer.

ABCD parameters of certain components such as shunt impedances and transmission lines can

be calculated in closed form given the physical properties of the components. Other components

commonly used in powerline networks, such as transformers and cap banks, that lack a physical

model at the frequency range of interest need to be characterized using vector network analyzers.

The overall approach of using TL modeling is illustrated in Fig. 4. A given powerline network,

with its electrical components, is mapped onto an equivalent topology of ABCD parameters (or

other network parameters). The equivalent ABCD between the desired TX-RX pair is computed

and used to calculate the transfer function for that link. As Fig. 4 shows, this approach is

generative since the topological aspects of the PLC-based network can be leveraged to compute

the transfer function between any TX-RX pair in the network. The IEEE P1901.2 standard

contributions given in [9] and [10] give a detailed analysis of using ABCD models to characterize

channel responses of field data for an outdoor NB-PLC system. They also provide measured

parameters for various components used in the field such as transformers and capacitor banks.

C. Time Varying Channels

Many electric devices connected to the power line network have a linear and time-invariant

current-voltage relationship. As a result, they can be characterized by a static impedance. However,

other devices exhibit nonlinear behavior that results in cyclic impedance properties with a period

typically equal to half the AC mains cycle [11]. If these devices dominate the system’s impedance,


100 200 300 400−90

−80

−70

−60

−50

−40

−30

Frequency (kHz)

Cha

nnel

Mag

nitu

de (d

B)

time t0time t0 + T/4

time t0 + T/2

0 10 20 30 40 50−200

0

200

time (msec)

0 10 20 30 40 500.2

0.40.60.8

1

time (msec)

AC V

olta

geno

rmal

ized

chan

nel n

orm

Charged Transformer

MV to LV Channel

Channel

Low Voltage

Medium Voltage

Fig. 5. The periodic time-varying channel of a charged transformer. The shape of the frequency response remains

relatively constant while the channel norm varies periodically at a period T = TAC/2. [Transformer image was

obtained from shutterstock.com]

the channel will exhibit a time-varying cyclic behavior of the same period. For example, the

channel across a charged transformer exhibits this periodic time-varying behavior which is shown

in Fig. 5. This figure suggests a simple approximation for modeling the periodic channel of the

transformer given by

hT (t) = α

[sin

(4π

TACt

)+ β

]h0 (t) (3)

where α is a gain constant, β is a positive offset, and h0 (t) is a common channel response.

More general periodically time-varying channel models have been proposed in [11] and [6],

while a discrete time-varying block channel model capturing this effect is proposed in [12]. On

the other hand, random switching, connecting, and disconnecting devices from the network result

in random load variations that lead to a form of time-selective fading.

IV. NOISE MODELS

Powerline noise is a significant factor that contributes to the hostile environment observed in

NB-PLC systems. This noise deviates significantly from the additive white Gaussian (AWGN)

assumption typically used to design and analyze communication systems. Typical sources of

PLC noise include: switching power supplies, silicon-controlled rectifiers, brush motors, dimmer

switches and industrial sources directly connected to the supply network [3], [11]. The resulting

noise is non-white with a time-varying spectral content that exhibits a 1/f -type decay due to the

decreasing concentration of noise sources with frequency and the used semiconductor devices.


Fig. 6. A noise trace captured at a medium and low voltage field sites with some possible noise sources [9]. The

noise exhibits cyclostationary features both in time and frequency domain. The trace can be divided into multiple

regions in which the noise can be approximated by a stationary process with a given power and spectral shape [15].

A. Periodic and Cyclostationary Noise

Many empirical results on PLC noise properties have been reported in [13]. The noise was

shown to have three main components: generalized background noise, periodic noise, and asyn-

chronous impulsive noise. However, these studies address the noise in the 0.2-20 MHz range;

thus, they are more applicable for broadband PLC systems.

More recently, studies targeted specifically for NB-PLC have shown that periodic noise and

cyclostationary noise are the dominant components of the additive noise in these systems [3],

[14]–[16]. A typical time-domain noise trace and a spectrogram of noise samples collected in

the field at medium and low voltage sites with some possible sources is given in Fig. 6. Both

traces, in addition to field data presented in the IEEE P1901.2 contribution given in [9], exhibit

cyclostationarity in both the time and frequency domain. In [14], the authors propose to use

a cyclostationary Gaussian model to characterize this aspect of the additive noise in NB-PLC

systems. This model captures the temporal characteristics of the noise by a Guassian process

with a time-varying variance that is periodic with a period equal to half the AC cycle; i.e.

n [k] ∼ N(0, σ2 [k]

)with σ2 [k] = σ2 [k +mT ] (4)

where T = 0.5TAC × FS , TAC is the duration of the AC cycle, FS is the sampling frequency,

and m ∈ Z. Spectral coloring is introduced by passing this noise through a linear time-invariant

shaping filter hs[k] with an exponentially decaying spectral density (see Fig. 7). A disadvantage


of this model is the complex parametric form used to characterize σ2 [k] that leads to a data-

intensive parameter estimation procedure. However, the primary disadvantage is that this model

decouples the frequency and temporal shaping. As a result, a single spectral shape is used to

shape the noise during each AC cycle which might not be appropriate when each region of the

AC cycle has a different spectral shape (see Fig. 6 and Fig. 7). Since the main cyclostationarity

captured is temporal, we refer to this model as the temporal cyclostationary model. In order to

capture the cyclostationarity in both the time and frequency domain, the authors in [9] and [15]

propose a novel cyclostationary model, which we refer to as spectro-temporal cyclostationary

model. This model divides each noise period into M regions during which the noise is stationary

(see Fig. 6). Each region is characterized by a spectral shape and a corresponding shaping filter.

Using this model, the PLC noise is modeled as the convolution of an AWGN signal s[k] with a

linear periodically time-varying system h[k, τ ] given by

n[k] =∑τ

h[k, τ ]s[τ ] =M∑i=1

1k∈Ri

∑τ

hi[τ ]s[τ ] (5)

where 1A is the indicator function, and h[k, τ ] =M∑i=1

hi[τ ]1k∈Rifor 0 ≤ k ≤ N − 1. This can

be implemented using a filter bank as shown in Fig. 7. The linear time-invariant filters hi[k]

correspond to spectral shaping filters for each region of the spectrogram in Fig. 6 and can be

estimated using spectrum estimation techniques [15]. A comparison between the two modeling

approaches is discussed in Fig. 7.

In addition to the cyclostationary noise, [16] shows that in the very low frequency (VLF),

below 10 kHz, there is an additional wide-sense periodic noise component. Presently, the VLF

range is extensively used by AMR providers. This periodic structure is exploited in [16] to filter

out this component and increase the post-processing SNR of these systems.

B. Uncoordinated Interference

Powerline cables constitute a shared medium over which PLC operates. Due to this shared

nature, PLC channels are potentially interference limited. As a result, recent increase in PLC stan-

dards and deployments has led to PLC coexistence issues [1], [3]. Interference from uncoordinated

devices can lead to significant reductions in data rates and affect overall reliability. Uncoordinated

transmissions could either result from other NB-PLC systems (deployed for other applications or

a competing standard) or from broadband technologies used for in-home networking due to lack


0 5 10 15 20 25 30−0.2

0

0.2

time (ms)

volta

ge (V

)

35

50

100

150

200

250

300

350

400

450

Freq

uenc

y (k

Hz)

50

100

150

200

250

300

350

400

450

50

100

150

200

250

300

350

400

450

dBV

/Hz0

.5

−140

−130

−120

−110

−100

−90

−80

−70

−60

0 5 10 15 20 25 30time (ms)

−0.2

0

0.2

0 5 10 15 20 25 30

time (ms)

−0.2

0

0.2

Noise Trace Captured at a Low Voltage Site Temporal Cyclostationary Model [13] Spectro-Temporal Cyclostationary Model [9,14]

Fig. 7. The two cyclostationary noise models for NB-PLC: i) the temporal cyclostationary model [14] shapes an

AWGN signal s[k] by the temporal function σ[k] and filters it through a single shaping filter hs[k], ii) the spectro-

temporal cyclostationary model [15] filters an AWGN signal s[k] by a filter bank of different shaping filters hi[k] each

corresponding to a stationary region of the noise period. Since the temporal cyclostationary model is constrained to

a single spectral shape, it overestimates the noise power spectral density in the 50 − 200 kHz region by as much as

50 dBV/Hz. The spectro-temporal model provides a closer fit to the measured noise spectrogram.

of isolation. A time division multiple access (TDMA) approach, called Inter-System Protocol

(ISP) [17], can provide coexistence between compliant standards such as IEEE 1901. An ISP

stand-alone coexistence scheme is provided by G.9972. As a result, technologies that choose to

implement it can coexist with IEEE 1901. However, there are already PLC technologies deployed

in the field that do not implement ISP, particularly for metering applications, that are expected to

have high deployment density. As a result, potential interference from these uncoordinated systems

would be treated as noise at the receiver. Since uncoordinated transmission can be random and

impulsive, it might not be fully captured by the noise models discussed above. This interference-

based noise will be similar to asynchronous impulsive noise investigated in [13], [18] and will

follow a Middleton Class-A or Gaussian mixture distribution.

In addition to uncoordinated PLC devices, there is interference from other uncoordinated man-

made technologies such as broadcast stations over medium- and short-wave broadcast bands [13].

This interference is manifested as a strong additive narrowband noise observed in many field data

such as the spectrogram of the medium voltage noise in Fig. 6 (at around 320kHz).


TABLE I

COMPARISON BETWEEN PLC AND WIRELESS ENVIRONMENTS

Wireless Narrowband PLC (3-500 kHz)

Time selectivity of the channel is due to the node

mobility.

Time selectivity results from random load variations

due random switching, in the power grid.

The time-varying stochastic nature captured by its

Doppler spectrum.

Time variation typically periodic with a period equal to

half the AC mains [11] with lognormal time-selective

fading [6].

Power decays as d−η/2 where d is the distance and η

is the propagation constant.

Power decays as e−α(f)d along a powerline wire and is

further attenuated when passing though transformers.

Additive noise is typically assumed stationary and

Gaussian.

Additive noise is non-Gaussian and impulsive with a

dominant cyclostationary component (see Section IV).

Dynamically changing propagation environment. Some level of determinism provided by the fixed grid

topology which can be exploited for deployment.

Typically interference limited. Potentially interference limited by uncoordinated users

due to increased deployment of different standards [1].

MIMO widely used such as in WiMAX and LTE. MIMO of order (# of wires - 1) is possible (such as

the G.9964 MIMO standard for BB-PLC).

Global synchronization across the network is difficult. The AC main signal can simplify synchronization.

V. PLC CHANNELS VS. WIRELESS CHANNELS

The wireless communications field has seen tremendous developments that have completely

transformed our way of life. Technologies such as digital modulation, digital signal processing

(DSP), Orthogonal Frequency Division Multiplexing (OFDM), wireless networking, and Multiple-

Input Multiple-Output (MIMO) systems have matured and became part of the standard commu-

nication toolset. The extent to which these tools can be leveraged for PLC depends largely on

PLC channel conditions. For comparison, we contrast the two environments in Table I.

VI. COMMUNICATION TECHNIQUES

Previous sections describe the channel and noise conditions that exist in a typical PLC channel.

Fig. 2 shows that the channel is typically better in the lower frequency ranges, while Fig. 6

suggests that the noise is higher in that frequency range. These conflicting quantities result in an


100 200 300 400

−80

−60

−40

−20

0

20

40

Frequency (kHz)S

NR

(dB

)

LV MV MV MVMV LV

Fig. 8. The combined effect of the channel and noise: the SNR varies significantly from one site to another [19],

[20]. It is not clear which band is the most suitable for transmission. In the MV→LV channel the higher frequency

band would be more appropriate while for the other two channels the lower bands look more promising.

ambiguity about what frequency subband is the most appropriate for transmission. This is illus-

trated in Fig. 8 where we show the signal to noise ratio (SNR) for various PLC channels. In this

section, we show how some of the widely used signal processing techniques for communications

can be used for PLC and where their application needs to be tailored specifically for the PLC

channel and the challenges it poses.

A. Multi-Carrier Modulation

Orthogonal frequency division multiplexing (OFDM) uses multiple orthogonal subcarriers to

transmit data over frequency selective channels such as in Fig. 2. These are transformed into a

collection of flat channels where each allows for a simple one-tap equalization. OFDM systems

can provide the robustness needed to deal with varying PLC channels through water-filling based

bit/power loading schemes. However, current narrowband PLC standards don’t employ these

algorithms due to their additional complexity. Instead, narrowband standards, such as G.hnem,

employ tone mapping, a simpler version of water-filling [21]. A tone map specifies which tones

will be loaded with bits versus which tones will be unloaded, or nulled, based on the SNR (see

Fig. 8). Currently, the same number of bits is assigned to each loaded tone; however, unequal

bit assignment is currently under investigation [21]. An additional benefit of OFDM is also

its improved resilience to impulsive noise over single-carrier systems [22]. At the receiver, the

discrete Fourier transform (DFT) spreads out the impulsive energy across all tones providing a


diversity effect that improves detection performance [22]–[24]. This is discussed in Section VI-C.

B. Coherent Vs. Non-Coherent Modulations

In an OFDM symbol, data can be modulated onto each subcarrier using coherent (e.g. PSK,

QAM) or non-coherent (e.g. DPSK) modulation schemes. Evaluation of the performance of

coherent vs. non-coherent detectors in various channel and noise conditions, as well as for

different application scenarios, still remains an active research area. In general, the choice between

coherent modulation and non-coherent modulation reflects complexity vs. performance tradeoffs.

Non-coherent modulation relieves the receiver from the computationally expensive channel es-

timation and carrier frequency offset correction. Proper selection of the window length ensures

that the channel phase distortion within it is approximately constant and thus is automatically

canceled when subtracting the phases of two consecutive symbols. Assuming perfect channel

estimation, the SNR gain of PSK over DPSK is generally 3 dB in AWGN. In non-Gaussian noise

environments, the SNR loss of DPSK increases over 3 dB as the noise becomes more impulsive

[25]. This is because the occurrence of high amplitude noise impulses can result in noisy phase

reference and severe error propagation. On the other hand, the performance of non-coherent coded

modulation can be improved by increasing the receiver complexity. In [26], the authors proposed

a non-coherent maximum likelihood sequence estimator (NMLSE) using overlapped observation

blocks. In AWGN, by increasing the observation length, the performance of the non-coherent

coded modulation with NMLSE converges to that of the coherent coded modulation with the

Viterbi decoder. However, such performance gain is significantly suppressed due to the presence

of impulsive noise [27].

C. Impulsive Noise Mitigation

In highly impulsive noise environments, the decrease of subchannel SNR can be too severe

to be mitigated by forward error correction (FEC) and interleaving. In this situation, denoising

algorithms can be used to remove the impulsive noise from the received signal before passing it

to the detector. These methods generally utilize the sparse structure of the impulsive noise. Two

compressed sensing (CS) algorithms were described in [23] and [24]. These methods utilize the

information contained in the null tones of an OFDM symbol to cancel out sufficiently sparse

but strong impulse noise. Recently, two sparse Bayesian learning (SBL) methods were proposed


for impulsive noise mitigation in OFDM systems [28]. The methods utilize information either

on the known subcarriers, such as null tones and pilots, to cancel out impulsive noise or on all

subcarriers to perform joint estimation-detection of impulse noise and data symbols. In addition,

these algorithms can be augmented using statistical models of the impulse noise to boost the

detection performance significantly over the CS methods.

VII. CURRENT STANDARDS

PRIME and G3 are the well-known industry-developed OFDM-based narrowband PLC stan-

dards. Both operate in the CENELEC-A (CEN A) band from 3 to 95 kHz (ITU-T G.9955/G.9956

and IEEE P1901.2 extend G3 to the FCC band from 159.4 to 478.1 kHz), and adopt non-

coherent modulation (i.e. DPSK) in combination with forward error correction (FEC) coding

(e.g. convolutional coding (Conv) and Reed-Solomon coding (RS)) to combat adverse channel

impairments and noise scenarios. Details about both standards and the comparisons between them

are well-described in [29] and [30]. PRIME and G3 PLC modems have been developed by several

companies, and have demonstrated success in field trials [31], [32].

To resolve the interoperability issues among the existing technologies, unified international

standards have been developed for the next generation narrowband PLC technology. The very

recently approved ITU-T Recommendations G.9955 and G.9956 specify the physical layer (PHY)

and the data link layer (DLL), respectively, of G.hnem and document PRIME and G3 to facilitate

the transient period [21]. Similarly, IEEE P1901.2 is being developed to provide advanced features

based on G3. Both these standards enable scalable data rates up to 500 kbps over a portion of

the CENELEC band (3–95 kHz in CEN A, 95–125 kHz in CEN B and 125–148.5 kHz in CEN

CD) and the entire US Federal Communications Commission (FCC) band (34.375–487.5 kHz).

Some PHY layer parameters of these two new standards are listed in Table II and compared to

those of PRIME and G3. Note that since the PHY specifications of PRIME and G3 are defined in

baseband real-valued settings, the FFT length is twice the number of subcarriers, whereas IEEE

P1901.2 and G.hnem are specified in complex OFDM settings, where the FFT length is equal

to the number of subcarriers. The maximum data rates take into account of cyclic prefix (CP),

forward error correction (FEC) coding, and the overhead of frame control headers, preambles,

channel estimation symbols and pilots.


TABLE II

PHYSICAL LAYER PARAMETERS OF PRIME, G3, IEEE P1901.2 AND G.HNEM.

Parameter PRIME G3 IEEE P1901.2 G.hnem

Frequency CEN A 42–89 kHz 35.9–90.6 kHz 35.9–90.6 kHz 35.9–90.6 kHz

Range FCC / 159.4–478.1 kHz 35.9–487.5 kHz 34.4–478.1 kHz

Sampling CEN A 250 kHz 400 kHz 400 kHz 200 kHz

Frequency FCC / 1.2 MHz 1.2 MHz 800 kHz

FFT LengthCEN A 512 256 256 128

FCC / 256 256 256

Cyclic Prefix CEN A 48 (192µs) 30 (75µs) 30 (75µs) 20/32 (100µs/160µs)

(Duration) FCC / 30 (25µs) 30 (25µs) 40/64 (50µs/80µs)

Window CEN A 0 8 8 8

Size FCC / 8 8 16

Effective CP CEN A 48 (192µs) 22 (55µs) 22 (55µs) 12/24 (60µs/120µs)

(Duration) FCC / 22 (18.3µs) 22 (18.3µs) 24/48 (30µs/60µs)

Subcarrier CEN A 488 Hz 1.5625 kHz 1.5625 kHz 1.5625 kHz

Spacing FCC / 4.6875 kHz 4.6875 kHz 3.125 kHz

OFDM CEN A 2240µs 695µs 695µs 700/760µs

Duration FCC / 231.7µs 231.7µs 350/380µs

Modulation4DPSK DPSK DPSK (QAM) QAM

M=2,4,8 M=2,4,8 M=2,4,8,16 M=2,4,8,16

FEC Conv (Optional) Conv+RS Conv+RS Conv+RS

Maximum CEN A 61.4/123 kbps5 45 kbps 52.3 kbps 101.3 kbps

Data Rate FCC / 207.6 kbps 203.2/207.6 kbps6 821.1 kbps

The comparison of modulation schemes among the standards in Table II reveals a new trend

that favors coherent over non-coherent modulation used in PRIME and G3. In particular, coherent

modulation is optional in IEEE P1901.2 and mandatory in G.hnem. The purpose is to guarantee

higher data rates and reliable communication [21], [27], and provide potential support for in-

home PLC applications. As discussed in the Section VI-B, coherent detection could significantly

improve receiver’s sensitivity and robustness, especially in the presence of impulsive noise.

However, it generally requires higher complexity for channel estimation. For example, in the

optional coherent modulation modes in IEEE P1901.2, two preamble symbols are added after

4For IEEE P1901.2, coherent modulation with BPSK/QPSK/8PSK/16-QAM is optional. M is the constellation size.5With the optional convolutional coding turned on and off, respectively.6For coherent and non-coherent modulations, respectively.


the frame control header (FCH) for channel estimation. An additional feature of IEEE P1901.2

is that, unlike G.hnem, it provides an optional adaptive multi-tone mask for the preamble and

header. This allows for PHY/MAC protocols that facilitate reliable communication for MV/LV

crossing in the US grids demonstrated in various field tests [33].

VIII. DEPLOYMENT AND CHALLENGES

The vast majority of today’s deployments are based on frequency shift keying (FSK) and

Spread-FSK as specified in the IEC 61334 standard. These low data rate NB-PLC technologies

provide 1.2-2.4 kbps with high robustness. However, as the data rate required by Smart Grid

applications increases, so does the interest in deploying high data rate OFDM-based PLC tech-

nologies such those discussed in Section VII. While the verdict on which standard will dominate

is still out, the likely answer will depend on the deployment environment, regulations, and the

specific application. Here we review some of the deployment aspects and challenges.

A. Deployment Enviroment

Topological characteristics of power grids have been studied in [1], [34]. Characteristics such

as connectivity, clustering, and average node degrees can provide a better understanding of the

power grid as a communication network. For example, [1] analyzes an MV rural network and

shows that due to its low connectivity it is prone to become disconnected under node failure.

More study cases across the world are necessary to gain a good statistical model for different

grids both on the MV and LV sites. This in turn will provide intuition about network design and

robustness (such as designing hybrid PLC/wireless networks). Without this statistical modeling

of world-wide grid statistics, we approach deployment from the perspective of meter density per

transformer and the deployment enviroment.

In rural deployments, the number of households per meter on the LV side of the MV/LV

transformer will typically be small (in the range of < 5− 10). In this case, it is not economical

to have a concentrator on the low voltage (LV) side of the transformer. Similarly, the desired

range for reaching the household meters can be large, on the order of several miles especially in

countries like USA and Europe. In [35], it was reported that with an attenuation of 3-5 dB per km

at ∼100 kHz and up to 6-8 dB per km in higher frequency range of ∼ 400 kHz, achieving these

long distances of tens of kms for rural applications may be challenging for PLC technologies


employing the 50-500 kHz frequencies. Typically, the UNB technologies such as TWACS can be

used for these deployments. For semi-rural applications in the USA, the number of LV meters

per transformer is still approximately 5-10, while the distances are smaller in the range of 3-

5 km with a higher meter density. In this case, technologies designed for crossing the MV/LV

transformers should be used for deployment. Field trials have confirmed that G3 is capable of

crossing the transformer in both directions [32]. Although awaiting empirical verification, by

design one would expect that P1901.2 and G.hnem to achieve this as well making them valid

deployment candidates. On the other hand, for urban deployments with very high user density,

the number of LV side nodes is larger in the range of 100 to also 1000 in Spain/France. In this

case, it is economical to have a concentrator on the low voltage side of the transformer and PLC

technologies like PRIME/G3/IEEE P1901.2/G.hnem can be used for deployment.

B. Throughput Requirements and Standards Coexistence

Rural applications, with a low household density per square km, typically require low data

rates in the order of few 100 bps. On the other hand, semi-rural and urban applications, due to

the higher density of households, require higher data rates in the tens of kbps range. In addition,

an accurate estimate of data rates needed for various Smart Grid application is still an open

problem and will need to take into account the high correlation in the collected data [1].

Also, as highlighted in Section IV-B, the coexistence of different PLC standards is a growing

problem. With the widespread adoption of PLC technology, interference from uncoordinated users

could prove to be a bottleneck in communication performance.

IX. CONCLUSION

In this article, we characterize the channel and the additive noise of a medium that was primarily

designed for one-directional flow of power at a main frequency of 50 or 60 Hz. NB-PLC systems

use some or all of the 3 to 500 kHz band. Power transmission lines exhibits attenuation, frequency

selectivity, and fading. Additional nonlinear and linear impairments are due to the transformer.

Cyclostationary noise forms the dominant component of noise in these systems. We conclude

by showing how to mitigate channel impairments and additive noise given models for them,

esp. in the context of current and emerging single carrier and multicarrier NB-PLC standards.

In general, PLC systems are attractive because they can be deployed over existing power lines.


NB-PLC systems, in particular, enable two-way communication links between customer sites

and a monitoring and control center at the local utility. Such links form a key component for

intelligently monitoring and controlling energy flows in the future Smart Grid, enabling improved

efficiency and reliability of power delivery.

ACKNOWLEDGMENT

The authors would like to express thanks for Aclara and Mr. Gordon Gregg for their support and

facilitation of the experiments where the data was collected. Furthermore, the authors extend their

thanks to Dr. Stefano Galli and the paper reviewers for their feedback that helped make this article

accurate and up-to-date with the latest standard developments. In addition, Mr. Nassar, Ms.Lin,

Mr. Mortazavi, and Prof. Evans were supported by the Semiconductor Research Corporation

under GRC ICSS Task ID 1836.063. Last, the authors would like to acknowledge the use of clip

art in the public domain from the OpenClipArtLibrary project in parts of Figures 1, 5, 6, and 7.

REFERENCES

[1] S. Galli, A. Scaglione, and Z. Wang, “For the grid and through the grid: The role of power line communications

in the smart grid,” Proc. IEEE, vol. 99, no. 6, pp. 998–1027, 2011.

[2] H. Hrasnica, A. Haidine, and R. Lehnert, Broadband Powerline Communications : Network Design, 2005.

[3] H. C. Ferreira, L. Lampe, J. Newbury, and T. G. Swart, Eds., Power Line Communications: Theory and

Applications for Narrowband and Broadbank Communications over Power Lines. Wiley, 2010.

[4] H. Philipps, “Modeling of powerline communication channels,” Proc. Int. Symp. Power Line Commun. and Its

Appl., pp. 14–21, 1999.

[5] M. Zimmermann and K. Dostert, “A multipath model for the powerline channel,” IEEE Trans. Commun., vol. 50,

no. 4, pp. 553–559, 2002.

[6] S. Galli, “A novel approach to the statistical modeling of wireline channels,” IEEE Trans. on Commun., vol. 59,

no. 5, pp. 1332–1345, 2011.

[7] T. Banwell and S. Galli, “A new approach to the modeling of the transfer function of the power line channel,”

Proc. Int. Symp. Power Line Commun. and Its Appl., pp. 319–324, 2001.

[8] O. Hooijen, “On the relation between network-topology and power line signal attenuation,” Proc. Int. Symp.

Power Line Commun. and Its Appl., pp. 45–56, 1998.

[9] A. Dabak, B. Varadrajan, I. H. Kim, M. Nassar, and G. Gregg, Appendix for noise channel modeling for IEEE

P1901.2, IEEE P1901.2 Std., June 2011, doc: 2wg-11-0134-05-PHM5.

[10] V. Borisov and P. Chiummiento, Impedance Variation Impairment on OFDM PLC PHY, IEEE P1901.2 Std.,

January 2011, doc: 2wg-11-0010-01-PHM5.


[11] F. Canete, J. Cortes, L. Diez, and J. Entrambasaguas, “Analysis of the cyclic short-term variation of indoor power

line channels,” IEEE J. Sel. Areas Commun., vol. 24, no. 7, pp. 1327–1338, 2006.

[12] S. Galli and A. Scaglione, “Discrete-time block models for transmission line channels: Static and doubly selective

cases,” IEEE Trans. Commun., September 2011, submitted, available online: http://arxiv.org/abs/1109.5382.

[13] M. Zimmermann and K. Dostert, “Analysis and modeling of impulsive noise in broad-band powerline commu-

nications,” IEEE Trans. Electromagn. Compat., vol. 44, no. 1, pp. 249–258, 2002.

[14] M. Katayama, T. Yamazato, and H. Okada, “A mathematical model of noise in narrowband power line

communication systems,” IEEE J. Sel. Areas in Commun., vol. 24, no. 7, pp. 1267–1276, 2006.

[15] M. Nassar, A. Dabak, I. H. Kim, T. Pande, and B. L. Evans, “Cyclostationary noise modeling in narrowband

powerline communication for smart grid applications,” Proc. IEEE Int. Conf. on Acoustics, Speech, and Signal

Process., pp. 3089–3092, March 2012.

[16] D. Rieken, “Periodic noise in very low frequency power-line communications,” Proc. IEEE Int. Symp. Power

Line Commun. and Its Appl., pp. 295–300, 2011.

[17] S. Galli, A. Kurobe, and M. Ohura, “The inter-PHY protocol (IPP): A simple coexistence protocol for shared

media,” Proc. IEEE Int. Symp. Power Line Commun. and Its Appl., pp. 194–200, 2009.

[18] M. Nassar, K. Gulati, Y. Mortazavi, and B. L. Evans, “Statistical modeling of asynchronous impulsive noise in

powerline communication networks,” Proc. IEEE Int. Global Commun. Conf., pp. 1–6, December 2011.

[19] D. Rieken, D. Langenberg, M. Walker, G. Gregg, I. H. Kim, A. Dabak, and T. Pande, Channel and Noise

Measurements In The US Rural Grid New Results With Improved Coupler and Measurement Procedure, IEEE

P1901.2 Std., April 2011, doc: 2wg-11-0086-00-PHM5.

[20] ——, Channel and Noise measurements, IEEE P1901.2 Std., April 2011, doc: 2wg-11-0124-00-PHM5.

[21] V. Oksman and J. Zhang, “G.HNEM: the new ITU-T standard on narrowband PLC technology,” IEEE Commun.

Mag., vol. 49, no. 12, pp. 36–44, 2011.

[22] J. Haring, Error Tolerant Communication over the Compound Channel. Shaker-Verlag, Aachen, 2002.

[23] G. Caire, T. Al-Naffouri, and A. Narayanan, “Impulse noise cancellation in OFDM: an application of compressed

sensing,” in Proc. IEEE Int. Sym. on Inf. Theory, 2008, pp. 1293–1297.

[24] L. Lampe, “Bursty impulse noise detection by compressed sensing,” Proc. IEEE Int. Symp. Power Line Commun.

and Its Appl., pp. 29–34, 2011.

[25] H. Huynh and M. Lecours, “Impulsive noise in noncoherent M-ary digital systems,” IEEE Trans. Commun.,

vol. 23, no. 2, pp. 246–252, 1975.

[26] D. Raphaeli, “Noncoherent coded modulation,” IEEE Trans. Commun., vol. 44, no. 2, pp. 172–183, 1996.

[27] D. Umehara, M. Kawai, and Y. Morihiro, “Performance analysis of noncoherent coded modulation for power

line communications,” Proc. Int. Symp. Power Line Commun. and Its Appl., pp. 291–298, 2001.

[28] J. Lin, M. Nassar, and B. L. Evans, “Non-parametric impulsive noise mitigation in OFDM systems using sparse

Bayesian learning,” in Proc. IEEE Int. Global Commun. Conf, 2011, pp. 1–5.

[29] M. Hoch, “Comparison of PLC G3 and PRIME,” Proc. IEEE Int. Symp. Power Line Commun. and Its Appl., pp.

165–169, 2011.


[30] I. H. Kim, B. Varadarajan, and A. Dabak, “Performance analysis and enhancements of narrowband OFDM

powerline communication systems,” in Proc. IEEE Int. Conf. on Smart Grid Commun., 2010, pp. 362–367.

[31] A. Aruzuaga, I. Berganza, A. Sendin, M. Sharma, and B. Varadarajan, “PRIME interoperability tests and results

from field,” in Proc. IEEE Int. Conf. on Smart Grid Commun., 2010, pp. 126–130.

[32] K. Razazian, M. Umari, A. Kamalizad, V. Loginov, and M. Navid, “G3-PLC specification for powerline

communication: Overview, system simulation and field trial results,” Proc. IEEE Int. Symp. Power Line Commun.

and Its Appl., pp. 313–318, 2010.

[33] B. Varadarajan, I. H. Kim, A. Dabak, D. Rieken, and G. Gregg, “Empirical measurements of the low-frequency

power-line communications channel in rural North America,” Proc. IEEE Int. Symp. Power Line Commun. and

Its Appl., pp. 463–467, 2011.

[34] Z. Wang, A. Scaglione, and R. Thomas, “Generating statistically correct random topologies for testing smart grid

communication and control networks,” IEEE Trans. Smart Grid, vol. 1, no. 1, pp. 28–39, 2010.

[35] A. Dabak, B. Varadarajan, I. H. Kim, V. Olson, and S. McHann, MV/LV communication modeling using S-

parameter/ABCD parameters, IEEE P1901.2 Std., March 2011, doc: 2wg-11-070-00-PHM5.

Marcel Nassar ([email protected]) received a B.E. degree in computer and communications engineering, with a

minor degree in mathematics, from the American University of Beirut in 2006, and an M.S. degree in electrical and

computer engineering from The University of Texas at Austin in 2008, where he is currently pursuing a PhD degree.

His interests lie at the intersection of signal processing and machine learning with application to statistical modeling

and mitigation of interference in communication systems.

Jing Lin ([email protected]) received a B.S. degree in electrical engineering in 2008 from Tsinghua University,

China, and an M.S. degree in electrical and computer engineering in 2010 from the University of Texas at Austin,

where she is currently pursuing a PhD degree. Her research has been focused on signal processing algorithms and

implementations for powerline communications.


Yousof Mortazavi ([email protected]) received his B.S. from the University of Tehran, Iran in 2005. In

2006, he joined the Electrical and Computer Engineering Department at The University of Texas at Austin and was

awarded the M.S. in 2008. Currently he is working towards his Ph.D. at the same department. His interests are signal

processing for communication systems and calibration and digital post-processing in mixed-signal integrated circuits,

specifically for data converters.

Anand Dabak ([email protected]) received the B.Tech. degree in electrical engineering from the Indian Institute of

Technology, Bombay, India in 1987 and the M.Sc. and Ph.D. degrees in EE, from Rice University, Houston, TX, in 1989

and 1992. He joined Texas Instruments Incorporated, Dallas, TX, in 1995 and has since then worked on the algorithm

issues related to wireless and power line communications (PLC). He has been involved in the standardization activity

and development in Bluetooth, ultra-wideband, digital video broadcasting-handheld and 3rd Generation Partnership

Project wideband code division multiple access, and long-term evolution. Currently he is TI -Fellow and working on

power line communication (PLC) and other problems in Smart Grid.

Il Han Kim ([email protected]) received the B.S. and M.S. degrees in electrical engineering from Korea Advanced

Institute of Science and Technology in 2002 and 2004, respectively, and Ph.D. degree in electrical engineering from

Purdue University in 2008. During 2004 and 2005, he was with the Electronics and Telecommunications Research

Institute, Daejeon, Republic of Korea. Since January 2009, he has been with the Systems and Applications R&D

Center, Texas Instruments Incorporated, where he has been working on power line communications. His research

interests include the analysis of various communication systems.

Brian L. Evans ([email protected]) received a B.S. degree in electrical engineering and computer science from

the Rose-Hulman Institute of Technology in 1987 and M.S. and Ph.D. degrees in electrical engineering from the Georgia

Institute of Technology in 1988 and 1993, respectively. From 1993 to 1996, he was a post-doctoral researcher at the

University of California, Berkeley. Since 1996, he has been a faculty member in electrical and computer engineering

at The University of Texas at Austin. He currently holds the Engineering Foundation Professorship. In 1997, he won

the U.S. NSF CAREER Award. He is a Fellow of the IEEE.

SUBMITTED TO IEEE SIGNAL PROCESSING …users.ece.utexas.edu/~bevans/papers/2012/powerline/... ·...

Documents

Transcript of SUBMITTED TO IEEE SIGNAL PROCESSING …users.ece.utexas.edu/~bevans/papers/2012/powerline/... ·...